The excess incidence of non-insulin-dependent diabetes mellitus noted among African Americans in the past two decades may be attributable to variations in the distribution of specific risk factors, or the impact of these risk factors may differ by ethnicity or sex. Over the 16 years (1971-1987) of the First National Health and Nutrition Examination Survey (NHANES I) Epidemiologic Follow-up Study, 880 incident cases of diabetes mellitus developed among 11,097 white and black participants who were between the ages of 25 and 70 years at baseline. There were substantial differences among the four race/sex groups with respect to age at baseline, as well as marked differences in the distribution of several major risk factors for diabetes, including obesity, subscapular and triceps skinfold thickness, blood pressure, income, activity, and educational level. The age-adjusted incidence of diabetes over the course of the study was 15.0% among black women, while it was 10.9% among black men. White women and men experienced similar, more moderate risks of 7.0% and 6.9%, respectively. The 100% excess risk among black women and the 50% excess among black men can in large measure explain the recent marked increase in diabetes rates in the black community. Furthermore, at nearly every level of obesity, blacks had a higher risk of diabetes than whites, suggesting that other factors contributed to risk. A significant interaction between race and body mass index (weight (kg)/height(m)2) was likewise demonstrated in multivariate analysis. Baseline age, race, body mass index, and ratio of subscapular skinfold to triceps skinfold were significantly related to incident diabetes, both overall and in separate models for men and women; in the entire cohort and in women alone, blood pressure, activity level, and education also contributed to risk. Other interactions were tested but were not found to be important. Despite sampling difficulties and inconsistencies in the data, the NHANES I Epidemiologic Follow-up Study provides evidence that the associations of anthropometric and sociodemographic variables with diabetes may vary among subgroups which have different mean levels and distributions of these risk factors.